What we did
We took a commercial dataset of every company that announced a funding round between 1 June and 31 July 2026, filtered it to those with a reachable website, and loaded each homepage twice within the same second: once in a real headless browser that executed JavaScript, and once with a plain HTTP fetch that did not.
The comparison between those two documents is the entire study. If a page is full of content when JavaScript runs and nearly empty when it does not, then every system that reads the web without executing JavaScript — ChatGPT, Claude, Perplexity, Google’s ad-quality crawler — receives the empty version.
1,403 domains were attempted and 1,348 returned a usable response. Median disclosed round in the sample: $6.6 million. Just under half were American, across 64 countries in total.
One in twelve serves almost nothing
117 of the 1,348 homepages met our definition of a shell: more than 600 characters of text when rendered, and under 25% of that surviving without JavaScript. That threshold is deliberately strict — the page has to be substantial when rendered and nearly empty when not.
The severity is the part worth sitting with. On a typical one of those 117, the rendered page carried 5,358 characters of text. The same URL fetched without JavaScript carried 55. Ninety of the 117 retained under 5% of their text.
That is not a degraded experience. It is an empty container. An assistant asked about one of those companies has nothing to read — not a thin page, no page.
The finding that surprised us
We expected the shells to cluster on cheap, old, neglected sites. They did the opposite.
Sites built on Wix produced zero shells out of 50. Squarespace, zero out of 41. Framer, 2.2%. WordPress, 3.3%. Shopify, 3.2%. Meanwhile Next.js came in at 8.5% — 21 of 247 sites — and HubSpot CMS at 10.0%. The 500 sites we could not attribute to any platform, which skew toward bespoke builds, were worst of all at 14.0%.
A company that raised millions and hired real engineers is more likely to be invisible to an assistant than a company paying sixteen dollars a month for a website builder.
The mechanism is not mysterious. A hosted builder emits finished HTML because it must — it was designed for people who will never open a terminal. A modern framework offers a choice between server rendering and client rendering, and client rendering is the path of least resistance. Next.js can render on the server; these 21 sites were configured not to.
It is not a money problem
The obvious objection is that shells belong to the smallest and youngest companies, and that the problem solves itself with scale. It does not.
Split by round size, the shell rate runs 9.4% under $1M, 9.2% between $1M and $5M, 9.4% between $5M and $20M, and 7.2% above $20M. Split by headcount it runs 8.8% at one to ten employees and 6.6% at 101 to 250. Nearly flat in both directions.
A company that raised more than twenty million dollars is only marginally less likely to be invisible than one that raised under a million. Whatever is driving this, it is not a shortage of budget or staff. It is a technical decision nobody revisited.
Twenty-eight are paying for clicks into a page the ad platform cannot read
Of the 117 shells, 28 carry Google Ads tags and 38 carry some paid pixel. Google evaluates landing pages to assess quality, and it evaluates them the same way every other crawler does. A page that returns 55 characters is being scored on those 55 characters while the advertiser pays for every click delivered to it.
This is the one finding where the cost is immediate and denominated in dollars rather than in future visibility. Across the whole sample, 327 companies were running Google Ads.
What we deliberately did not conclude
Three findings survived a first pass and were deleted after audit, and we publish them because a study that reports only what confirmed its thesis is not worth reading.
We nearly wrote that certain companies’ Meta pixels fire a page view and nothing else. A pixel fires a lead event when somebody submits a form, and a crawler that loads one page and submits nothing can never observe it. That claim would have been false on 98 companies.
We nearly wrote that some accounts have no conversion tracking. Google Ads conversion tags fire on the thank-you page, not the homepage — unobservable by this method.
And we nearly overstated how many sites measure nothing. Consent banners hold every tag until a click nobody made; several analytics vendors were missing from our detection patterns entirely; and hosted builders ship their own measurement. Each correction cut the count, from 327 to 304 to 274.
We also do not claim any specific crawler saw this. Our fetch used a normal browser user-agent, not GPTBot’s. What we measured is that the content is unavailable without JavaScript execution — which is the condition those crawlers operate under.
Check your own in a minute
The test costs nothing. Fetch your homepage without executing JavaScript and read what comes back. If your headline, your services and your copy are not in the response, then assistants, ad-quality crawlers and link previews are all working from whatever little is there.
We built a free tool that does it for you at revioconsulting.com/tools/ai-visibility — paste a domain and it shows the readable text, the title, the H1, the meta description and any tags visible in the source. No sign-up.
The full study, including method, sample composition and the findings we deleted, is at revioconsulting.com/research/funded-site-audit. Aggregate results only — no company is identified.
If your site turns out to be one of these, the fix is usually server rendering or a prerender step rather than a rebuild. We did exactly that on this site, and the difference was zero to roughly 1,600 words of crawlable text on the homepage.